End-to-end brain causal network construction method based on graph neural network

Abstract

The invention discloses an end-to-end brain causal network construction method based on a graph neural network, and belongs to the field of electroencephalogram information processing. According to the method, a multilayer perceptron with adjacent k layers of feature fusion is designed for multi-dimensional feature extraction, and a drawing neural network is further designed for direct mining of the brain causal relationship. Then, a multivariate sequence with real electroencephalogram signal characteristics and causal supervision information of the multivariate sequence are obtained through a vector autoregression model, and a neural network model is trained through a supervision method; and based on the trained neural network model, mining of the causal relationship of the electroencephalogram data and construction of the causal network can be realized. Compared with a representative method of a traditional method, Granger causal analysis comparative research proves that the method has remarkable advantages in the aspect of capturing the causal network topological structure and causal relationship strength under the condition of low signal-to-noise ratio. According to the method, a new perspective is provided for breaking through traditional model-driven hypothesis constraints and directly mining a deep brain causal network mechanism in a data-driven mode.

Description

technical field [0001] The invention belongs to the field of EEG information processing, in particular to an end-to-end brain causal network construction method based on a graph neural network. Background technique [0002] Brain network is of great significance to the study of the interaction between various brain regions, and the causal network describes the causal influence characteristics between various brain regions. In order to describe the causal relationship of brain activity, a variety of methods have been proposed, which can be roughly divided into a series of model-driven methods represented by Granger causality (GCA) and a small number of non-parametric estimation methods. However, the series of methods based on model-driven assumptions depend on the reliability of the model assumptions, and are limited by the discrepancy between the model-driven assumptions and reality, making it difficult to describe the nature of the causal network mechanism of the brain. Re...

AI Technical Summary

Problems solved by technology

And because most of them need to be trained for a single sample, the model cannot adapt to the complex noise environment of the brain signal, and thus has limitations in the generalization of the model, computational cost, etc.

Images